PERFORMANCE ANALYSIS OF BRAZILIAN HEDGE FUNDS

Transcription

1 PERFORMANCE ANALYSIS OF BRAZILIAN HEDGE FUNDS GUSTAVO A. JORDÃO INSPER INSTITUTE OF EDUCATION AND RESEARCH MARCELO L. DE MOURA INSPER INSTITUTE OF EDUCATION AND RESEARCH Abstract This paper analyzes Brazilian hedge funds, a market of $160 billion of assets under management in September By assessing four asset pricing models the study tests the claim that hedge funds can produce abnormal returns, gain market momentum (present market timing) and have low correlation with market risk. To create the estimates, we utilized a database of 1,673 funds, with monthly data from January 2000 to August 2009, provided by the Brazilian Association of Financial and Capital Markets Institutions (ANBIMA). Reported results indicate that this market presented a small number of funds with the ability to obtain abnormal returns (5% on average) and a small number of administrators with the ability to adjust market timing (3.7% on average). However, this market presented better results with respect to risk, with approximately 35% of funds showing no correlation with the market. Besides the reported performance results, the study provides two additional important contributions to the literature. First, the paper investigates a representative emerging market and second, in the light of the very strong financial crisis in the second semester of 2008, the paper studies hedge fund performance in the scenarios of high volatility and financial stress (June 2008 to November 2008) and also in the scenario of very rapid recovery rally (December 2008 to August 2009). Keywords: multimarket funds, hedge funds, calculation of alpha, market timing, CAPM, fund performance assessment, Brazilian hedge funds. JEL: G11 Author for correspondence: Rua Quatá, 300. São Paulo/SP, BRAZIL, ; marcelom@insper.org.br ; tel.: (The former Ibmec São Paulo has been renamed Insper Institute of Education and Research ). 1

2 1 - Introduction With around $400 billion in assets under management in the U.S. in 2004 (see Capocci and Hubner, 2004), and estimates for 2009 of about $1.25 trillion, 1 hedge funds emerged around 1950 and went on to gain prominence in the 1990s. Despite being present in many countries, such funds differ along several factors, such as minimal levels of capital investment, the administrator s obligatory participation in the fund, recommendations of investment values and management strategies. In Brazil, funds that adopt some of the main strategies of the hedge fund industry are known as multimarket funds (fundos multimercado). These funds do not require the participation of the administrator s capital and do not restrict access to customers with little capital (some funds allow the participation of small investors). In September 2009, these funds registered a net worth of US$160 billion, representing about 8% of the total net worth of all Brazilian funds, just below the proportion constituted by fixed income funds, Brazilian interbank deposit interest rate (DI) referenced funds and fixed income pension securities (30%, 17% and 9%, respectively). 2 Nonetheless, several studies show that this industry has superior returns (see Oberuc, 1994, and Hennessee, 1994) when compared to mutual funds, including those mentioned above (fixed income, Brazilian DI referenced funds and the fixed income pension funds). The objective of this study is to analyze the performance of Brazilian hedge funds between January 2000 and February To do so, we estimate several common performance indicators by analyzing the main asset pricing models. This article fits within the literature assessing the performance of hedge funds, and is similar to the work of Steri et al. (2008), Capocci and Hubner (2004) and Liang (1999). The remainder of the paper is divided into five sections. The next section (Section 2) provides a brief literature review. Section 3 explains the factor models used and the selected performance indicators, Section 4 describes the data, and Section 5 presents the results. The final section concludes as well as discusses limitations and possible extensions of this study. 2 Literature on performance analysis models The performance of hedge funds is a recent theme in the economics literature. Early works on the subject date back to the 1980s. The methodologies employed vary from the simplest CAPM model, to the multi-factorial models of Fama and French (1993) and Carhart (1997) to the panel model of Steri et al. (2008). The results have been varied; some studies have found that hedge funds perform better than mutual funds, but worse when compared to benchmarks (see Ackermann et al., 1999 and Liang, 1999), while others find that funds show return persistence, that is, a consistently above the average performance (see Agarwal and Naik, 2000). Next, we present a brief discussion of the main models developed in this field, organized in chronological order. The pioneering model, widely considered a benchmark in the area of finance, was developed by Sharpe (1964) and Lintner (1965), and is known as the Capital Asset Pricing Model (CAPM). This model determines that the return of a portfolio of high-risk assets minus the return of the risk-free rate is a function of the market risk premium. The alpha of the 1 Source: Wall Street Journal, Source: Brazilian Association of Financial and Capital Markets Institutions (ANBIMA),

3 Sharpe (1964) model (1964), called Jensen's alpha (1968) is interpreted as an evaluation of undervalued or overvalued assets. The beta is interpreted as the portfolio risk before market oscillations, known as systematic risk. Fama and French (1993) developed an extension to CAPM, related to the returns of portfolios with two additional factors counted as hypothetical portfolios, following a selection criterion of assets that the authors believed to have ties with the returns. The first factor, called SMB, measures the company size effect, represented by the series of returns of a hypothetical portfolio, with zero investment, purchased in shares of "small" companies and sold in shares of "large" companies. The second factor, called HML, measures the market value effect, represented by a hypothetical portfolio, also with zero investment, with positions purchased in stocks with high book-to-market equity (by the market value net worth) and sold in shares with low book-to-market equity. It is noteworthy that, in Brazil, the market value index by the company net worth is commonly used. However the authors considered the opposite calculation, that is, net worth by the market value (net assets by shareholder equity). Following the line of Fama and French (1993), Carhart (1997) added the factor he called the momentum effect, which considers the purchase of winning assets and the sale of losing assets. Although his article focuses on the returns of mutual funds and not of hedge funds, Carhart (1997) concluded that the costs of transactions and load fees (costs of brokers for the purchase and sale of assets) have a negative impact on the performance of these funds. Ackermann et al. (1999) developed an analysis that differed from the others, proposing a regression of the Sharpe ratio for each fund by variables that represent their own strategies and characteristics, such as performance rates, management fees and age. The model is different because it relates the returns of the funds to their intrinsic characteristics. The funds, each with their rates and strategies, have returns that can be justified by these characteristics, which were not considered in the models of Fama and French (1993) and Carhart (1997). The authors' main finding is in regard to the major positive correlation between the performance rate and risk-adjusted return, the same conclusion reached by Liang (1999). His results show that the average Sharpe ratio of hedge funds is 21% higher than that of mutual funds; nevertheless, these funds fail to consistently outperform the market when calculating the risk-adjusted return (Sharpe ratio). Liang (1999), contributed to two distinct models. The first of these refers to a regression of returns on market indices, and the second, to the regression of average fund returns on their own characteristics, a model similar to that of Ackermann et al. (1999). The conclusion of Ackermann et al. (1999) work shows little correlation of returns with market indexes, and positive relationships with some fund characteristics, such as the lockup period, the performance rates and assets under administration. In contrast, there is a variable referring to fund age that produces negative impact on returns. Agarwal and Naik (2000) estimated a model similar to the first model of Liang (1999). However, when using different market indexes, they concluded that different fund strategies are exposed to different markets. The authors conclude that hedge fund returns exceed the market benchmark by 6% to 15%. 3

4 Edwards and Caglayan (2001) produced a peculiar model, in which a regression of the fund returns on certain market indexes is first performed, as in the case of Fama and French (1993) and Carhart (1997) models. Then, the values of unexplained returns (alpha values) are obtained in order to perform a regression on the different strategies adopted by funds. The authors concluded, like the others previously mentioned, that there is a positive relationship between the performance rates and funds returns, although the conclusion refers to abnormal returns. Capocci and Hubner (2004) proposed a model that can be considered an extension of the Carhart (1997), Fama and French (1993) and Agarwal and Naik (2002) models. The authors considered the relationship of the funds to the markets of emerging countries, since many funds invest in these countries. This model has been proven to be superior to those developed by Sharpe (1964), Fama and French (1993), Carhart (1997) and Agarwal and Naik (2002), because it better explains variations in hedge funds returns. The authors concluded that 25% of the funds analyzed presented significant return excesses. In the analysis performed by sub-periods, results indicated the constant ability of the management to achieve superior performance in the market. Do et al. (2005) evaluated the Australian market using a specification similar to that of Capocci and Hubner (2004) and also proposed a second model using another measure of performance, the modified Sharpe ratio. This ratio takes into account the non-normality of fund returns. In their conclusions, the authors found a worsening of the Sharpe ratio index with the new methodology (using the modified index); funds with positive results had less positive rates using the modified Sharpe ratio, and funds with negative Sharpe ratio indexes presented even worse indexes with this methodology. Further, Do et al. (2005) and Amin and Kat (2003) treated specifications taking into account the non-normality and the non-linearity of returns, as the authors found that hedge fund returns do not have normality and linearity as mutual funds do. The authors concluded that hedge funds are ineffective as single investments; however, the addition of a hedge fund to a portfolio presents improvements to the portfolio. Steri et al. (2008), using more advanced econometric methods, developed a study of the hedge funds market in Italy. The authors adopted a panel analysis to assess performance, using both temporal data of market indexes and cross-sectional data of individual characteristics of funds. Despite a discussion of possible inconsistencies in the use of the Sharpe ratio as a performance indicator, this study has several advantages due to its panel analysis. The authors concluded that the American securities market, as well as performance fees and redemption periods, produce a negative effect on the performance of hedge funds. The Brazilian investment fund literature, still incipient on this subject, analyzes the Brazilian funds market from the perspectives of the market timing ability of administrators (see Brito, 2003), of the analysis of Sharpe and Sortino ratios (see Fonseca et al., 2007) and of the persistence of fund performance (see Xavier, 2008). Brito (2003) analyzed the funds market with active management and proposed a new index (Brito s Skill Index) to evaluate the performance and market timing of administrators. They found that few managers have significant market forecasting abilities. Fonseca et al. (2007) divided the market between fixed income funds and variable income funds (equity funds) and evaluated performance according to the Sharpe and Sortino ratios. They concluded that, in terms of returns, variable income funds proved to be better 4

5 than fixed income funds. However, in the risk/return ratio, the fixed income funds presented better results. According to the authors, this is because the country has high interest rates. Xavier (2008), in addition to evaluating the performance persistence of multimarket funds with variable income and leverage, also separated funds into three categories whose characteristics relate to the volatility of returns, the nominal return and the size of the fund. In his work, performance is analyzed according to the Sharpe ratio, and his findings showed evidence of performance persistence. Nevertheless, when divided into the three categories, results diverge. To summarize, the literature analyzing fund performance, although extensive, tends to focus on specific indicators. Consequently, the findings vary and strongly depend on the chosen indicators. As a contribution to this body of literature, this paper investigates, for the Brazilian market, not just one specific indicator, but several aspects and indicators assessed in the literature, such as the Jensen's alpha for testing abnormal returns, the market beta (see Agarwal and Naik, 2000) for testing the ability to mitigate market risk, as well as market timing performance indicators. In order to do that, we use an extensive set of factor models that we describe in the next section. 3 - Performance indicators Sharpe's (1964) CAPM predicts that an asset excess return will be determined by the correlation it has with the market excess return. It is represented mathematically by: where: R is portfolio return of asset i; i R R ( R R ) u (1) i f i i M f i R is return on a risk-free short-term asset; is the f intercept; is the slope; is return on the market portfolio; and is the random error factor. The model relates the expected return of a given risky asset by the risk-free rate and the market risk premium. Several hypotheses are assumed to validate the model: the inability of an individual investor to affect the price with his negotiations, the planning of a maintenance period of an identical investment for all investors; unlimited opportunities to take credit at a risk-free rate, in the absence of transaction costs; rational optimization of the average and variance of all investors; and homogeneous expectations. Even if some of the assumptions of the model may not be true in practical terms, the CAPM is still widely used in the calculation of abnormal returns and risk. The advantage of this model is its simplicity, in both theoretical (assessment) and practical terms. Its assessments provide a variety of information, including the verification of the supposed ability of an administrator to obtain abnormal returns, a positive i, not explained by the market systematic risk, a non-zero. Market timing is defined as the ability of fund administrators to anticipate market oscillations, so that they can benefit from the situation, taking positions with greater systematic risk when the market is high or less systematic risk when the market is low. It is calculated using a model similar to the CAPM, but takes into account a factor that computes the ability to make such predictions. The equation for market timing was proposed by Treynor and Mazuy (1996), who calculated the coefficient in the following regression: i 5

6 2 R R ( R R ) ( R R ) u i f i i M f i M f i (2) where is the factor that calculates market timing. When the coefficient is significant and i positive, the fund manager was successful in predicting market oscillations to increase the return of the fund. The estimation of this model in this paper will use one additional index. To calculate Treynor alpha and beta indexes, we will use the coefficients of the CAPM model in its simplest version, not the coefficients of the Treynor and Mazuy model (1996). Fama and French (1993) proposed a more sophisticated CAPM model to obtain a better explanation of returns. In their model, two factors were added: the first concerns the size of companies, while the second refers to the relationship between companies market value and equity. Therefore, the equation estimated in the model of Fama and French (1993) is as follows: R R ( R R ) SMB HML u (3) i f i i M f i i i where SMB is a series of a hypothetical portfolio (the difference between portfolio returns on small stocks and portfolio returns on large stocks); and HML is the return difference of portfolios with high book-to-market equity (equity value by the market value of the fund) and returns with low book-to-market equity. Carhart (1997) estimated a model similar to that of Fama and French (1993), in which a variable called momentum effect is added. The author performs the regression of excess return of funds in the following equation: R R ( R R ) SMB HML PR1YR u (4) i f i i M f i i i i where PR1YR is the momentum effect. As treated by Fama and French (1993), the above model proposes the following as explanatory variables: the excess return of the market; the difference between the portfolio return on small stocks and the portfolio return on large stocks (SMB); the difference in portfolio returns with high book-to-market equity and with low book-to-market equity (HML); and, finally, the momentum effect (PR1YR). This concerns the difference between the return of a hypothetical portfolio that includes companies with the highest returns over 11 months and the returns of a similar portfolio that comprises companies that have obtained the worst returns in that same time period. For each of the above models, Equations (1) to (4), the intercept, known as Jensen's alpha, will measure the administrator s ability index. This index, represented by the coefficient, quantifies the abnormal returns earned by the fund, since the return obtained by the administrator is not explained by any exposure to risk factors in the models CAPM and CAPM with market timing, or of Fama and French (1993) and Carhart (1997). Such returns are interpreted as resulting from the fund administrator s ability. The beta index, calculated by the CAPM model, represents the fund's systematic risk i. A fund that is close to zero has little risk on the market, that is, the market can oscillate both positively and negatively, and the fund will not feel major effects. This index is relevant because it quantifies the market risk carried by each fund. The allocation of resources among 6

7 various assets available on the market helps the administrator eliminate the specific risk of each asset. However, deciding the right time to buy and sell assets is not as simple as portfolio diversification. It is necessary that the fund managers predict the market oscillations in order to buy assets low and sell them high. The estimation of the CAPM model with market timing provides us with a coefficient that relates the manager's ability to anticipate the market and therefore benefit from the situation. This index is obtained by the regression Equation (2). When the coefficient of this equation is significant, the manager is successful; otherwise, the manager did not achieve adequate market timing. 4 - Analysis and description of the data We utilized a database of 1,673 funds, with monthly data from January 2000 to August 2009, provided by the Brazilian Association of Financial and Capital Markets Institutions (ANBIMA - Associação Brasileira das Entidades dos Mercados Financeiro e de Capitais). In this work, we included all funds considered as multimarket fund in ANBIMA s classification as representative of the Brazilian hedge funds. Accordingly to ANBIMA, those funds adopt some of the main strategies used by the called hedge funds in other countries. From the ANBIMA database we selected active and extinct funds between January 2000 and August Only the funds that had at least 48 monthly observations available were selected, resulting in a sample of 1,673 funds, 1,084 active and 589 inactive. 3 Data pertaining to the benchmark (returns of the Brazilian stock market, Bovespa index) and the risk-free rate (CDI Interbank Deposits certificates) were gathered using the Thomson/Reuters DataStream. The period under analysis began in January 2000 and ended in August The descriptive analysis of the data and models to be estimated considers three periods; these are the total life-span of the funds, i.e., the period from the creation of the fund until its end or until August, 2009 (for still active funds); the period that includes the worst losses of the Bovespa index given the global financial crisis in 2008, June, 2008 to November, 2008, with an accumulated loss of 47% in the Bovespa index; and, lastly, the recovery period from December, 2008 to August, 2009, during which the Bovespa index rallied 48%. Table 1 below presents the descriptive statistics of the hedge funds for each period. Considering all of the available observations of the funds (first period), the average monthly return is 1.27%, with a standard deviation of 1.36%. This is an average return is slightly lower than the benchmark stock market index (Ibovespa), but it has a standard deviation that significantly lower, 1.19% against 8.10%. Comparing minimum and maximum returns for all the three selected periods for Funds and Ibovespa clearly demonstrate the higher risk involved in the Brazilian stock markets. [INSERT TABLE 1 ABOUT HERE] 3 It is noteworthy that some funds have been located with a monthly return observation that is unreasonably high, with returns above 100% in a month. As outliers, these observations were eliminated. 7

8 Interestingly, investments in the short term risk free interest rate 4, CDI, earned a slightly higher rate of return of 1.27% than the average hedge fund. This is possibly due to the fact that Brazilian short term interest rate was kept in very high values, mainly in early 2000 s in order to keep inflation under control. Also, in order to take advantage of those high interest rates, most of the multimarket funds were heavily invested in short term government bonds earning the CDI rate. During the most acute period of the financial crisis in 2008, hedge funds, as well as the Ibovespa, performed relatively worse. Surprisingly, hedge funds yielded positive returns on average, probably due to their supposedly high allocation on short-term government bonds. On the other hand, during the recovery period, Ibovespa performed much better than the hedge funds industry. Analyzing funds individually, we can gain much more information regarding hedge funds average returns and standard deviations by period. In figure 1 we display histograms for average returns and standard deviations for the three selected periods in our sample. Table 2 reports the comparison of the hedge funds average, standard deviations and accumulated returns with the Ibovespa and CDI. Looking at average returns in figure 1 for the three selected periods, it is clear that the life-span, crisis and recovery period have very distinct patterns, with approximately symmetric distribution in the full sample, positively skewed distributions in the crisis period and negatively skewed distribution during the recovery period. Standard deviations increase in the crisis period and decreases in the recovery period samples relatively to the full sample period, [INSERT FIGURE 1 AND TABLE 2 ABOUT HERE] Accordingly to table 2, for the total life-span of the hedge funds during the period of January 2000 to August 2009, only 18.2% and 31.7% of them had, respectively, greater average and accumulated returns than the Ibovespa. Those numbers increases to about half of the sample when we compare hedge funds returns with the short term risk-free rate, CDI. Although in the crisis period almost all of the hedge funds had better returns than the Ibovespa, most of them also fell behind the market during the recovery period. In summary, all this empirical evidence suggests that Brazilian hedge funds are less volatile than the stock market but correlated to it. If we analyze the total life-span of the hedge funds, just a few, 31.7%, were able to beat the stock market and just about half were able to get greater performance than the high nominal short term interest rates in Brazil. Those preliminary results give us a hint that just a few hedge funds will get high performance indicators, a result we will further investigate in section 5. In order to complete the description of the data, Table 3 below presents summary statistics of the explanatory variables for the proposed models. They are: market excess return (Ibovespa return minus the CDI return), HML factor, SMB factor and PR1YR factor. It can be observed from the table that the average of the factors proposed by Fama and French (1993), excess return, SMB and HML, was positive during the period January 2000 to 4 The CDI closely is an interbank deposit rate that follows very closely the short term policy interest rate in Brazil, denominated Selic rate. The Selic indexes the most part of government bonds, known as LTN (Letras do Tesouro Nacional). Market convention, however, uses CDI instead of the Selic rate as the short-term interest rate benchmark. Therefore, we can assume that short term investments in LTN government bonds earn the risk-free rate given by the CDI. 8

9 February This result is in accordance with the findings of Fama and French (1993), with a premium of 0.19% per month of the SMB factor and a premium of 0.18% per month of the HML factor. The PR1YR factor must also have a positive result, given that this is a hypothetical portfolio of the difference between a portfolio that reached the best results in the last 11 months and a portfolio that reached its worst results in the same period. The total number of observations of all factors is 116, in the period between January 2000 and August The majority of these series do not present normality, observed by the zero probability associated with the Jarque-Bera test, with the exception the market excess returns. [INSERT TABLE 3 ABOUT HERE] 5 Results The CAPM and CAPM with market timing models, as well as those of Fama and French (1993) and Carhart (1997) described above, if estimated by OLS (Ordinary Least Squares), can generate consistent estimators, which are nevertheless inefficient if there is heteroskedasticity or autocorrelation in the errors. In such cases, it is not possible to make statistical inferences. When we think about returns on the stock market in emerging countries like Brazil, we should consider some unique characteristics, such as heteroskedasticity and autocorrelation, thereby creating the specific hypothesis necessary to estimate models by OLS. White (1980) proposed a solution by estimating a matrix of consistent covariance. This estimate, known as least squares robust estimation, makes the estimators consistent and efficient in cases of heteroskedasticity. However, as stated previously, in emerging markets we also find problems of autocorrelation in errors, making the White's estimation still inefficient. In view of this problem, Newey and West (1987) proposed a more general estimator for the covariance matrix, correcting the lack of homoskedasticity and the lack of zero autocorrelation. Therefore, all of the proposed models are estimated by OLS corrected by the covariance matrix of Newey and West (1987). Table 4 shows the distributions of Jensen's alpha abnormal returns, the beta coefficients, accounting for the systematic market risk and the market timing coefficients. Most of the hedge funds under analysis were unable to obtain performance superior to the benchmark, since the number of funds with positive and significant alphas (fourth column of Panel A) represents only 3.7% to 7.8% of the total funds, depending on the model selected, with an average across models of 5%. Surprisingly, the percentage of hedge funds with statistically significant and negative alphas is much higher, ranging from 8.3% to 13.5%, 12% on average. Given that these hedge funds are intended to achieve abnormal returns and to outperform the market and the risk-free rate, we could infer that their objectives were not achieved, as only about 5% on average have achieved their purpose and 12% have negative alphas, for the majority of them, 83%, alphas are null. Figure 3 shows histograms of the results of alpha coefficients generated by the four estimated models. As expected, given the results discussed previously, the distribution has higher mass on the negative values as the alphas have slightly negative means. 9

10 [INSERT TABLE 4 ABOUT HERE] [INSERT FIGURES 2, 3 AND 4 ABOUT HERE] To analyze the management of funds under the risk perspective, we should infer with regard to the beta coefficients generated in the models. Panel B in table 4 presents the results in all models for both significant and non-significant coefficients. The result that most interests us is the non-significance of these coefficients, since a zero beta is relative to the effective hedging of the market systematic risk. Notice that the number of funds whose coefficient is statistically zero falls in the interval from 538 to 562 funds (fifith column). This interval corresponds to an average of 33% of funds with the capability to minimize market risk. By detailing the values of the betas, we note that on average, of the significant coefficients, 96% correspond to positive values. We conclude that, of those funds in which systemic risk is present, most funds follow market oscillations, i.e., when the market rises the funds tend to be more profitable and the fund becomes less profitable as the market falls. As might be expected based on the results of the analysis, the histograms of the beta coefficients shown in Figure 4 are all positively skewed. [INSERT FIGURE 4 HERE] As previously mentioned, the CAPM with market timing and the Cahart (1997) models attempt to capture the administrator s ability to forecast the market. According to equation (2) it does not matter if the oscillation is positive or negative (the market excess return is squared) since the manager can take advantage of both the rises and falls in the market. In the Cahart (1997) model, the ability is measured as by taking advantage of being long on the high return stocks and long on the low return stocks. Therefore, to confirm the existence of market timing, the coefficient i of equations (2) and the coefficient i of equation (4) must be positive and significant. Panel C in table 4 illustrates the results of these estimations. Only 3.5% of the hedge funds have market timing ability accordingly to CAPM with market timing model. The evidence of this skill is approximately the same for the Cahart (1997) model with only 3.9% of funds presenting positive and statistically significant coefficients. Actually, for the CAPM with market timming, a significant part of hedge funds lose money with higher market oscillations, 18.5% of the hedge funds present a negative and significant market timing coefficient. Figure 4 presents the distribution of results generated by the estimation of market timing coefficients. 6 - Conclusions, limitations and possible extensions The present study analyzed the performance of Brazilian hedge funds. Considered a market that is still incipient in Brazil, these funds have gained in popularity since the 1990s. In the U.S., where there are already considerable financial movements and significant market value, this market is still much smaller than the mutual funds market. They differ from mutual funds because they are based on active management with the possibility of investing 10

11 in several markets and with a certain degree of leverage. These funds are an object of study in the literature, since they are capable of outperforming the market indexes or any defined benchmarks. Based on the availability of data on the Brazilian market, the work presented here was based on four models of performance analysis. These were the CAPM model, the CAPM with market timing model, the Fama and French (1993) model and the Carhart (1997) model. The performance evaluation in this work focused on three performance indicator variables, they are: abnormal returns (alpha), systematic risks (betas), market timing indicators. The abnormal returns are related to the ability of administrators to obtain returns that are not explained by the market risk of the fund. Beta relates to systematic risk: the closer beta is to zero, the more efficient the fund is in regard to mitigate market risk. Market timing measures the administrators ability to predict market oscillations: when the coefficient is positive and significant, the manager was successful in his forecasts over the analyzed period. The results were not the most favorable for the Brazilian hedge funds when we analyzed the ability of administrators to obtain returns unexplained by systematic risk (Jensen's alpha), since only 3.7% to 7.8% of the funds presented positive and significant alpha coefficients. These results are substantially lower than those obtained by Capocci and Hubner (2004) in the U.S. market. Performance in the market timing ability were not striking either: on average, only 3.7% of hedge funds administrators (estimated by the CAPM with market timing and the Cahart (1997) models) could effectively predict the market successfully in order to aggregate greater returns to the fund. On the subject of managing risk, the results were slightly more favorable, as many of the betas estimated in all models were not significant. About 33% of the funds had zero systematic risk. This paper provides an analysis of the performance of Brazilian multimarket funds in the perspective of the financial crisis in 2008 and the subsequent recovery of the markets from December 2008 to August Besides the reported performance results, the study provides two additional important contributions to the hedge funds performance literature. First, the paper investigates hedge funds in a representative emerging market with a representative size of $160 billion of AUM at the end of 2009, with relatively scarce literature. Second, in the light of the very strong financial crisis in the second semester of 2008, the paper studies hedge fund performance in the scenarios of high volatility and financial stress (June 2008 to November 2008) and also in the scenario of very rapid recovery rally (December 2008 to August 2009). Clearly the study has some important limitations that should be addressed in future work. First, the Brazilian market classification of multimarket funds (fundos multimercados) as hedge funds is subjected to some criticism. Although they have the characteristic of being permitted by regulators to undertake a wider range of investment and use unconventional strategies, some of them are heavily invested in government bonds taking advantage of high interest rates that prevailed in Brazil form 2000 t Therefore, a more narrow definition of hedge funds should be used. Second, the issue of performance persistence was not addressed. Finally, another possible extension should investigate the economic value if identifying higher alpha, lower beta and market timing funds by building portfolios based on those performance indicators. 11

12 References ACKERMANN, Carl; MCENALLY, Richard; RAVENSCRAFT, David. The Performance of Hedge Funds: Risk, Return, and Incentives. In: The Journal of Finance, vol. 54, n o 3, p AGARWAL, Vikas. Intertemporal Variation in the Performance of Hedge Funds Employing a Contingent-Claim-Based Benchmark. In: EFMA 2001, Lugano Meetings. Disponível em: SSRN: ou DOI: /ssrn AGARWAL, Vikas; NAIK, Narayan Y. On Taking the Alternative Route: the Risks, Rewards, and Performance Persistence of Hedge Funds. In: Journal of Alternative Investments, vol. 2, pp AMIM, Gaurav S.; KAT, Harry M. Hedge Fund Performance : Do the Money Machines Really Add Value? In: The Journal of Financial and Quantitative Analysis, vol. 38, n o 2, pp BODIE, Zvi. KANE, Alex. MARCUS, Alan J. Fundamentos de Investimentos. São Paulo: Editora Bookman, pp BRITO, Ney Roberto O. Avaliação de desempenho e Market Timing: o índice de habilidade. In: Revista Brasileira de Finanças, v. 1, nº 1, junho de pp CAPOCCI, D.; HUBNER, G. Analysis of Hedge Funds Performance. In: Journal of Empirical Finance, vol. 11, p CARHART, M. M. On persistence in mutual fund performance. In: Journal of Finance, vol. 52, pp DO, V., FAFF, R.; WICKRAMANAYAKE, J. An empirical analysis of Hedge Funds Performance: The case of Australian Hedge Funds industry. In: Journal of Multinational Financial Management, vol. 15, p EDWARDS, F.; CAGLAYAN, M. O. Hedge Fund Performance and Managers Skills. In: Journal of Futures Markets, vol. 21, pp ELING, M., Does hedge funds performance persist? Overview and Empirical Evidence. In: Working Paper. Saint Gallen: University of Saint Gallen, FAMA, E. F., FRENCH, A. K. R. Common Risk Factors in the Returns on Stocks and Bonds. In: Journal of Financial Economics, vol. 33, p FAMA, E. F.; FRENCH, A. K. R. Multifactor Explanations of Asset Pricing Anomalies. In: Journal of Finance, vol. 51, pp FONSECA, Nelson; BRESSAN, Aureliano A; IQUIAPAZA, Robert A.; GUERRA, João Paulo. Análise do desempenho recente de fundos de investimento no Brasil. In: Contab. Vista & Rev., vol. 18, nº 1, jan./mar pp

13 GREGORIOU, G. N.; GUEYIE, J. P. Risk-Adjusted Performance of Hedge Funds Using a Modified Sharpe Ratio. In: Journal of Wealth Management, vol. 6, pp HENNESSEE, E. L. The Republic New York Securities Quarterly Hedge Fund Review, março de JENSEN, M. C. The Performance of Mutual Funds in the Period In: Journal of Finance, vol. 23, pp LIANG, Bing. On the Performance of Hedge Funds. In: Financial Analysts Journal, vol. 55, n o 4, pp LINTNER, J. The Valuation of Risk Assets and the Selection of Risky Investments in Stock Portfolio and Capital Budgets. In: Review of Economics and Estatistics, vol. 47, pp NEWEY, Whitney K.; WEST, Kenneth D. A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. In: The Econometric Society, Econometrica, vol. 55, n o 3, maio de pp OBERUC, Richard. Performance of Hedge Funds Relative to Traditional Investiments. In: Working Paper, LaPorte Asset Allocation System, SHARPE, W. F. Capital Asset Prices: a Theory of Market Equilibrium under Conditions of Risk. In: Journal of Finance, vol. 19, pp STERI, R.; GIORGINO, M.; VIVIANI, D. The Italian Hedge Fund Industry: an Empirical Analysis of Performance and Persistence. In: Journal of Multinational Financial Management, p. 17. XAVIER, Antonio Luiz B. Persistência de performance: fundos multimercados com renda variável e alavancagem. Rio de Janeiro: Faculdade Ibmec, WHITE, Halbert. A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. In: The Econometric Society, Econometrica, vol. 48, n o 4, maio pp

14 TABLES Table 1 - Descriptive statistics of the Brazilian hedge funds, stock markets index (Ibovespa) and the short term risk-free interest rate (CDI) Fund life-span Jan/00 - Aug/09 Periods Financial crisis Jun/08 - Nov/08 Recovery Dec/08 - Aug/09 Statistic Asset Mean Funds Ibovespa CDI Stand. Funds Dev. Ibovespa CDI Median Funds Ibovespa CDI Min. Funds return Ibovespa CDI Max. Funds return Ibovespa CDI Note: All values are in percentage points and represent monthly discrete returns. Funds correspond to Brazilian hedge funds (denominated multimarket funds) obtained from Brazilian Association of Financial and Capital Markets Institutions (ANBIMA - Associação Brasileira das Entidades dos Mercados Financeiro e de Capitais) database. Only the funds that had at least 48 monthly observations available were selected, resulting in a sample of 1,673 funds, 1,084 active and 589 inactive. The series of returns composes the total life-span of the funds and may include funds with observations for the entire period of January 2000 to August 2009, as well as funds with only 48 observations in that interval. Data pertaining to the benchmark Ibovespa (returns of the Brazilian stock market, Bovespa, index) and the risk-free rate (CDI interbank deposits certificates) were gathered using the Thomson/Reuters DataStream database. The period under analysis began in January 2000 and ended in August 2009.The maximum and minimum monthly returns consider the average of these extremes for all funds in all analyzed periods. 14

15 Table 2 Brazilian hedge funds returns comparison with the stock markets index (Ibovespa) and the short term risk-free interest rate (CDI). Panel A - Mean returns Period # of funds Greater than Bovespa Greater than CDI Fund life-span (Jan/00 - Aug/09) 1, (18.2%) 898 (53.7%) Financial crisis (Jun/08 - Nov/08) 1, (99.6%) 268 (22.2%) Recovery (Dec/08 - Aug/09) 1, (1.1%) 898 (80.7%) Panel B - Standard deviations Period # of funds Greater than Bovespa Greater than CDI Fund life-span (Jan/00 - Aug/09) 1, (1.3%) 1669 (99.8%) Financial crisis (Jun/08 - Nov/08) 1, (1.0%) 1187 (99.7%) Recovery (Dec/08 - Aug/09) 1,096 9 (0.8%) 1030 (94.0%) Panel C - Accumulated returns Period # of funds Greater than Bovespa Greater than CDI Fund life-span (Jan/00 - Aug/09) 1, (31.7%) 880 (52.6%) Financial crisis (Jun/08 - Nov/08) 1, (99.7%) 266 (22.1%) Recovery (Dec/08 - Aug/09) 1, (1.0%) 897 (80.6%) Note: This table displays the number of funds with average returns and standard deviations greater than the Bovespa and the CDI indices presented within the same time-period for all investment alternatives. The series of returns composes the total life-span of the funds and may include funds with observations for the entire period of January 2000 to August 2009, as well as funds with only 48 observations in that interval. 15

16 Table 3 - Descriptive statistics of market factors explanatory variables. Factors Statistic Excess Return HML SMB PR1YR Mean Median Maximum Minimum Standard Deviation Asymmetry Kurtosis Jarque - Bera Probability Number of obs Note: Excess returns are the difference of the Brazilian stock market index (Ibovespa ) return and the short term risk-free interest rate (CDI ). The factors HML (high minus low) and SMB (small minus big) for the Brazilian market were computed using the methodology of Fama and French (1993) and the momentum effect (PR1YR) was computed using the methodology of Carhart (1997). 16

17 Table 4 - Results of the intercepts (alphas), systematic risk factors (betas) and market timing coefficients of the estimated models. Panel A - Alpha distribution All coefficients Only statistically significant at 5% Statistically = 0 Models CAPM (45.8%) (54.2%) + 83 (5.0%) (12.2%) 0 1,386 (82.8%) CAPM with Market timming 910 (54.4%) 763 (45.6%) 131 (7.8%) 138 (8.3%) 1,404 (83.9%) Fama & French(1993) 684 (40.9%) 989 (59.2%) 66 (3.9%) 223 (13.3%) 1,384 (82.7%) Carhart (1997) 651 (38.9%) 1,022 (61.1%) 62 (3.7%) 226 (13.5%) 1,385 (82.8%) Panel B - Beta distribution All coefficients Only statistically significant at 5% Statistically = 0 Models CAPM + 1,493 (89.3%) (10.8%) + 1,069 (63.9%) - 42 (2.5%) (33.6%) CAPM with Market timming 1,487 (88.9%) 186 (11.1%) 1,066 (63.8%) 43 (2.6%) 564 (33.7%) Fama & French(1993) 1,499 (89.7%) 174 (10.4%) 1,077 (64.4%) 45 (2.7%) 551 (32.9%) Carhart (1997) 1,501 (89.8%) 172 (10.3%) 1,093 (65.4%) 42 (2.5%) 538 (32.2%) Panel C - Market timming coefficient distribution All coefficients Only statistically significant at 5% Statistically = 0 Models CAPM com Market timming (35.5%) - 1,079 (64.5%) + 59 (3.5%) (18.5%) + 1,304 (77.9%) Carhart (1997) 958 (57.3%) 715 (42.7%) 66 (3.9%) 35 (2.1%) 1,572 (94.0%) Note: The estimated models match the time series regressions according to the specifications presented in Section 2. The series of returns composes the total life-span of the funds and may include funds with observations for the entire period of January 2000 to August 2009, as well as funds with only 48 observations in that interval. The level of significance was 5%. The percentages are in relation to the total number of funds, 1,

18 FIGURES Figure 1 - Relative frequency distributions of Brazilian hedge fund returns, Brazilian stock market (Ibovespa) and Brazilian short-term risk free rate (CDI) for three selected periods. Relative Frequency Relative Frequency Relative Frequency Average Returns - Full Sample Average Returns - Crisis Period Average Returns - Recovery Period Standard Deviation of Returns - Full Sample Standard Deviation of Returns - Crisis Period Standard Deviation of Returns - Recovery Period Ibovespa average (solid line) CDI average (dashed) 18

19 Figure 2 Histograms of the alpha coefficients of all estimated models..20 Alpha - CAPM.20 Alpha - CAPM with Market Timming Relative Frequency Alpha - Fama and French (1993).20 Alpha - Cahart (1997) Relative Frequency

20 Figure 3 - Histograms of the market excess returns beta coefficients of all estimated models Beta - CAPM Beta - CAPM with Market Timming Relative Frequency Beta - Fama and French (1993) Beta - Cahart (1997) Relative Frequency Figure 4 - Histograms of the coefficients of market timing Market timming coefficient - CAPM Market timming coefficient - Cahart (1997) Relative Frequency